Portfolio AI Ops
Deploy and measure AI across portfolio companies with a Brier-scored forecast ledger. Proof, not pilots.
Book a 15-min intro All 9 systems
Portfolio AI Ops —
Ledger
Calibration
Portfolio
IC Memo
fleet: idle
Initiative Company Forecast Due Value at stake Spend Outcome Brier

Brier per forecast = (p − outcome)². 0 is perfect, 0.25 is what you get for saying 50/50, 1.0 is confidently wrong. Open forecasts are not scored until their due date resolves.

Calibration
Whether the numbers mean anything. A forecaster who says 70% should be right about 70% of the time. This page is the audit trail that proves it — or proves it does not.

Reliability diagram n = 0 resolved

perfect calibration observed point size ∝ number of forecasts in bin

Brier score by cohort lower is better

Grouped by the half-year the forecast was opened. A downward slope is the only real evidence of learning.

Murphy decomposition

Every Brier score splits into exactly three parts. Reliability is calibration error and you want it near zero. Resolution is how far your forecasts move away from the base rate — it is the part that carries information, and you want it large. Uncertainty is fixed by the base rate of the questions themselves and no forecaster can change it. The identity below is checked numerically on every render.

Score by category

Score by portfolio company

Portfolio
Deployment and measured return per company. Spend is counted on every deployment attempted. Realised value is counted only where the forecast resolved YES.
Investment committee memo
Generated from the ledger as it stands right now. The numbers are computed locally and are not editable. Optionally send the computed figures to the local GPU fleet to draft the narrative around them.

Memo

Narrative drafting

Sends only the computed aggregate figures. Requires the LAN/Tailscale fleet to be reachable from this browser. If it is not, the deterministic memo above still works — it is generated entirely in-page.

Method note

Value at stake is the annualised P&L effect underwritten at the time the forecast was opened, agreed with the portfolio company CFO. It is not re-forecast after the fact. Spend is fully loaded deployment cost. A forecast that resolves NO forfeits its value but keeps its spend. That asymmetry is deliberate and is what an LP would insist on.

New forecast

It has to be settleable by a third party without argument. A named metric, a threshold, and a date.